Paper2Agent
A system proposed by a research team at Stanford University that automatically encapsulates academic paper methods as callable AI agents. By organizing papers, code, and data and connecting them to a chat agent, it enables interactive reuse of paper methods.
Timeline (last 90 days)
The Paper2Agent-related work proposed by a Stanford University research team was published in Nature under the title "Reimagining Research Papers as Interactive and Reliable AI Agents."
The core idea of Paper2Agent is to organize papers, supplementary materials, code, data and analysis pipelines into queryable materials and runnable tools, then connect them to a chat Agent, so that researchers can both ask about methodological details in natural language and have the Agent execute analyses step by step.
On a test of 300 tutorial questions, Paper2Agent achieved 91.2% accuracy, higher than the 80.3% achieved when a chat Agent directly accesses the corresponding code repository.
Paper2Agent's server contains three types of resources: tools for executable methods, static materials such as papers/code/data, and prompt templates.
Paper2Agent uses the Model Context Protocol (MCP) to specify how the chat Agent discovers materials, understands tool descriptions and initiates calls, giving the methods of different papers a unified invocation approach.
Paper2Agent automatically verifies each generated tool using examples from the original repository; the verification dimensions include whether the target files are generated, whether numerical differences are within tolerance, and whether charts correspond to the original figures, and failed verification triggers continued diagnosis and repair.
Automatic verification presupposes that the repository comes with reproducible tutorials or examples; if the original tutorial depends on specific hardware, private data, or unfixed random seeds, the reference outputs may be unobtainable, and such papers are difficult to bring into the tooling pipeline.
The team turned AlphaGenome into an Agent, generating 22 tools, all of which passed automatic verification, in about 45 minutes at a cost of about 14 US dollars.
The Scanpy Agent was run on four public single-cell datasets, and the cell and gene counts after quality control as well as the marker genes of each cell cluster were identical to the results researchers obtained using the same set of parameters.
Among 100 computational biology papers, Paper2Agent successfully turned 74 into usable Agents, proposing 599 tools in total, of which 593 passed verification.
3 more timeline events
All Facts (13)
The above evaluations cover only tasks defined by the papers (tutorial questions test execution fidelity, new questions test the ability to flexibly combine tools) and cannot represent the method's general capability across all research tasks.
50%UnverifiedThe core idea of Paper2Agent is to organize papers, supplementary materials, code, data and analysis pipelines into queryable materials and runnable tools, then connect them to a chat Agent, so that researchers can both ask about methodological details in natural language and have the Agent execute analyses step by step.
50%UnverifiedOn a test of 300 tutorial questions, Paper2Agent achieved 91.2% accuracy, higher than the 80.3% achieved when a chat Agent directly accesses the corresponding code repository.
50%UnverifiedPaper2Agent's server contains three types of resources: tools for executable methods, static materials such as papers/code/data, and prompt templates.
50%UnverifiedPaper2Agent uses the Model Context Protocol (MCP) to specify how the chat Agent discovers materials, understands tool descriptions and initiates calls, giving the methods of different papers a unified invocation approach.
50%UnverifiedPaper2Agent automatically verifies each generated tool using examples from the original repository; the verification dimensions include whether the target files are generated, whether numerical differences are within tolerance, and whether charts correspond to the original figures, and failed verification triggers continued diagnosis and repair.
50%UnverifiedAutomatic verification presupposes that the repository comes with reproducible tutorials or examples; if the original tutorial depends on specific hardware, private data, or unfixed random seeds, the reference outputs may be unobtainable, and such papers are difficult to bring into the tooling pipeline.
50%UnverifiedThe team turned AlphaGenome into an Agent, generating 22 tools, all of which passed automatic verification, in about 45 minutes at a cost of about 14 US dollars.
50%UnverifiedThe Scanpy Agent was run on four public single-cell datasets, and the cell and gene counts after quality control as well as the marker genes of each cell cluster were identical to the results researchers obtained using the same set of parameters.
50%UnverifiedAmong 100 computational biology papers, Paper2Agent successfully turned 74 into usable Agents, proposing 599 tools in total, of which 593 passed verification.
50%UnverifiedOn 42 tasks from 10 non-biological-computing papers, Paper2Agent achieved 98.1% accuracy.
50%UnverifiedFor 26 data- and discovery-oriented papers, Paper2Agent's resource layer answered 100 comprehensive questions with 89% accuracy, higher than the 82% baseline of browsing the web.
50%UnverifiedThe Paper2Agent-related work proposed by a Stanford University research team was published in Nature under the title "Reimagining Research Papers as Interactive and Reliable AI Agents."
50%